Unlocking AI's Hidden Connections With Graphs
summary
Claim: vector-only RAG misses multi-hop links and causes subtle hallucinations — an agent approves a $25,000 credit increase for a customer whose employer is on a sanctions list because the relationship is implicit not explicit. Solution: combine Neo4j knowledge graphs with embeddings-as-node-properties, run graph traversals plus algorithms like FastRP, k-NN, Louvain and PageRank, and record a context graph decision trace so the system can show its work and address 95% no measurable ROI findings from the MIT Nanda Initiative.
AI-written summary · about
chapters (12)
- 0:00Why AI Agents Make Bad Decisions
- 1:23A Better Data Architecture for AI
- 2:22The Hidden Cost of Broken Connections
- 4:00When Hallucinations Affect Real Decisions
- 5:00The Compliance Agent Failure
- 6:55Why Most AI Projects Don't Deliver ROI
- 8:02Knowledge Graphs vs Vector Databases
- 10:34What Is a Knowledge Graph?
- 14:14The Missing Dimension in RAG
- 15:25Context Graphs and Decision Tracing
- 18:01Demo: Explaining an AI Decision
- 22:11Making AI Auditable and Trustworthy
related talks
27mWhen Vectors Aren’t Enough: Unlocking AI’s Hidden Connections with Graphs
36mAgentic GraphRAG: AI’s Logical Edge
30mAgentic RAG with Knowledge Graphs Explained
18mConnecting the Dots with Context Graphs
28mData-Empowered AI Assistants
24mAgentic GraphRAG: AI’s Logical Edge
35mGraphRag in Action: Revolutionizing Fraud Detection Techniques
39mEnhancing RAG with Neo4j Knowledge Graph